Computer-assisted determination of left ventricular endocardial borders reduces variability in the echocardiographic assessment of ejection fraction

Machine Learning-Enabled Fully Automated Assessment of Left Ventricular Volume, Ejection Fraction and Strain: Experience in Pediatric and Young Adult Echocardiography

BACKGROUND

Left ventricular (LV) volumes, ejection fraction (EF), and myocardial strain have been shown to be predictive of clinical and subclinical heart disease. Automation of LV functional assessment overcomes difficult technical challenges and complexities. We sought to assess whether a fully automated assessment of LV function could be reliably used in children and young adults.

METHODS

Fifty normal volunteers (22/28, female/male) were prospectively recruited for research echocardiography. LV volumes, EF, and strain were measured both manually and automatically. An experienced sonographer performed all the manual analysis and recorded the analysis timing. The fully automated analyses were accomplished by 5 groups of observers with different knowledge and medical background. AutoLV and AutoSTRAIN (TomTec) were employed for the fully automated LV analysis. The LV volumes, EF, strain, and analysis time were compared between manual and automated methods, and among the 5 groups of observers.

RESULTS

Software-determined endocardial border detection was achievable in all subjects. The analysis times of the experienced sonographer were significantly shorter for AutoLV and AutoSTRAIN than manual analyses (both p < 0.001). Strong correlations were seen between conventional EF and AutoLV (r = 0.8373), and between conventional three view global longitudinal strain (GLS) and AutoSTRAIN (r = 0.9766). The volumes from AutoLV and three view GLS from AutoSTRAIN had strong correlations among different observers regardless of level of expertise. EF from AutoLV analysis had moderately strong correlations among different observers.

CONCLUSION

Automated pediatric LV analysis is feasible in normal hearts. Machine learning-enabled image analysis saves time and produces results that are comparable to traditional methods.

Fully Automated Mouse Echocardiography Analysis Using Deep Convolutional Neural Networks

Echocardiography (echo) is a translationally relevant ultrasound imaging modality widely used to assess cardiac structure and function in preclinical models of heart failure (HF) during research and drug development. Though echo is a very valuable tool, the image analysis is a time consuming, resource demanding process, and is susceptible to inter-reader variability. Recent advancements in deep learning have enabled researchers to automate image processing and reduce analysis time and inter-reader variability in the field of medical imaging. In the present study, we developed a fully automated tool - Mouse Echo Neural Net (MENN) - for the analysis of both long axis brightness (B)-mode and short axis motion (M)-mode images of the left ventricle. MENN is a series of fully convolutional neural networks that were trained and validated using manually segmented B-mode and M-mode echo images of the left ventricle. The segmented images were then used to compute cardiac structural and functional metrics. The performance of MENN was further validated in two preclinical models of HF. MENN achieved excellent correlations (Pearson's r = 0.85 to 0.99) and good to excellent agreement between automated and manual analyses. Further inter-reader variability analysis showed that MENN has better agreements with an expert analyst than both a trained analyst and a novice. Notably, the use of MENN reduced manual analysis time by >92%. In conclusion, we developed an automated echocardiography analysis tool that allows for fast and accurate analysis of B-mode and M-mode mouse echo data and mitigates the issue of inter-reader variability in manual analysis.

Timely-Automatic Procedure for Estimating the Endocardial Limits of the Left Ventricle Assessed Echocardiographically in Clinical Practice

In this paper, we propose an analytical rapid method to estimate the endocardial borders of the left ventricular walls on echocardiographic images for prospective clinical integration. The procedure was created as a diagnostic support tool for the clinician and it is based on the use of the anisotropic generalized Hough transform. Its application is guided by a Gabor-like filtering for the approximate delimitation of the region of interest without the need for computing further anatomical characteristics. The algorithm is applying directly a deformable template on the predetermined filtered region and therefore it is responsive and straightforward implementable. For accuracy considerations, we have employed a support vector machine classifier to determine the confidence level of the automated marking. The clinical tests were performed at the Cardiology Clinic of the County Emergency Hospital Timisoara and they improved the physicians perception in more than 50% of the cases. The report is concluded with medical discussions.

Challenges With Left Ventricular Functional Parameters: The Pediatric Heart Network Normal Echocardiogram Database

BACKGROUND

The reliability of left ventricular (LV) systolic functional indices calculated from blinded echocardiographic measurements of LV size has not been tested in a large cohort of healthy children. The objective of this study was to estimate interobserver variability in standard measurements of LV size and systolic function in children with normal cardiac anatomy and qualitatively normal function.

METHODS

The Pediatric Heart Network Normal Echocardiogram Database collected normal echocardiograms from healthy children ≤18 years old distributed equally by age, gender, and race. A core lab used two-dimensional echocardiograms to measure LV dimensions from which a separate data coordinating center calculated LV volumes and systolic functional indices. To evaluate interobserver variability, two independent expert pediatric echocardiographic observers remeasured LV dimensions on a subset of studies, while blinded to calculated volumes and functional indices.

RESULTS

Of 3,215 subjects with measurable images, 552 (17%) had a calculated LV shortening fraction (SF) < 25% and/or LV ejection fraction (EF) < 50%; the subjects were significantly younger and smaller than those with normal values. When the core lab and independent observer measurements were compared, individual LV size parameter intraclass correlation coefficients were high (0.81-0.99), indicating high reproducibility. The intraclass correlation coefficients were lower for SF (0.24) and EF (0.56). Comparing reviewers, 40/56 (71%) of those with an abnormal SF and 36/104 (35%) of those with a normal SF based on core lab measurements were calculated as abnormal from at least one independent observer. In contrast, an abnormal EF was less commonly calculated from the independent observers' repeat measures; only 9/47 (19%) of those with an abnormal EF and 8/113 (7%) of those with a normal EF based on core lab measurements were calculated as abnormal by at least one independent observer.

CONCLUSIONS

Although blinded measurements of LV size show good reproducibility in healthy children, subsequently calculated LV functional indices reveal significant variability despite qualitatively normal systolic function. This suggests that, in clinical practice, abnormal SF/EF values may result in repeat measures of LV size to match the subjective assessment of function. Abnormal LV functional indices were more prevalent in younger, smaller children.

Accurate assessment of LV function using the first automated 2D-border detection algorithm for small animals - evaluation and application to models of LV dysfunction

Echocardiography is the most commonly applied technique for non-invasive assessment of cardiac function in small animals. Manual tracing of endocardial borders is time consuming and varies with operator experience. Therefore, we aimed to evaluate a novel automated two-dimensional software algorithm (Auto2DE) for small animals and compare it to the standard use of manual 2D-echocardiographic assessment (2DE). We hypothesized that novel Auto2DE will provide rapid and robust data sets, which are in agreement with manually assessed data of animals.2DE and Auto2DE were carried out using a high-resolution imaging-system for small animals. First, validation cohorts of mouse and rat cine loops were used to compare Auto2DE against 2DE. These data were stratified for image quality by a blinded expert in small animal imaging. Second, we evaluated 2DE and Auto2DE in four mouse models and four rat models with different cardiac pathologies.Automated assessment of LV function by 2DE was faster than conventional 2DE analysis and independent of operator experience levels. The accuracy of Auto2DE-assessed data in healthy mice was dependent on cine loop quality, with excellent agreement between Auto2DE and 2DE in cine loops with adequate quality. Auto2DE allowed for valid detection of impaired cardiac function in animal models with pronounced cardiac phenotypes, but yielded poor performance in diabetic animal models independent of image quality.Auto2DE represents a novel automated analysis tool for rapid assessment of LV function, which is suitable for data acquisition in studies with good and very good echocardiographic image quality, but presents systematic problems in specific pathologies.

Assessment of left ventricular ejection fraction in critically ill patients at the time of speckle tracking echocardiography: intensivists in training for echocardiography versus experienced operators

BACKGROUND

The biplane Simpson's method is considered the gold standard to assess and monitor left ventricular (LV) ejection fraction (EF) in critically ill patients. Recently, a new semi-automatic technique based on speckle tracking echocardiography called "Auto-EF" has been introduced. We compared LVEF values obtained with biplane Simpson's method and Auto-EF by two groups of operators: trainee echocardiography intensivists and experienced echocardiographers.

METHODS

A standard transthoracic echocardiography was performed on 37 patients. According to image quality 29 patients were selected. Each inexperienced and experienced operator executed an off-line analysis using both Simpson's method and Auto-EF. LVEF obtained by the two groups of operators were then compared.

RESULTS

EF values assessed with Simpson's method showed a moderate correlation (r=0.70, P<0.01) between inexperienced and experienced operators. The Bland-Altman analysis showed a mean bias of 0.3% with limits of agreement (LoA) from -24.4 to +25.1%. Values obtained with Auto-EF showed a good correlation (r=0.94, P<0.01) with a mean bias of 0.2% and LoA from -10.1 to +10.4%.

CONCLUSIONS

Due to its semiautomatic nature, for inexpert operators Auto-EF seems more reproducible than the traditional Simpson's method for monitoring left ventricular function in critically ill patients.

Reproducibility of Left Ventricular Dimension Versus Area Versus Volume Measurements in Pediatric Patients With Dilated Cardiomyopathy

BACKGROUND

Multiple echocardiographic methods are used to measure left ventricular size and function. Clinical management is based on individual evaluations and longitudinal trends. The Pediatric Heart Network VVV study (Ventricular Volume Variability) in pediatric patients with dilated cardiomyopathy has reported reproducibility of several of these measures, and how disease state and number of beats impact their reproducibility. In this study, we investigated the impact of observer and sonographer variation on reproducibility of dimension, area, and volume methods to determine the best method for both individual and sequential evaluations.

METHODS AND RESULTS

In 8 centers, echocardiograms were obtained on 169 patients prospectively. During the same visit, 2 different sonographers acquired the same imaging protocol on each patient. Each acquisition was analyzed by 2 different observers; first observer analyzed the first acquisition twice. Intraobserver, interobserver, interacquisition, and interobserver-acquisition (different observers and different acquisition) reproducibility were assessed on measurements of left ventricular end-diastolic dimension, area, and volume. Left ventricular shortening fraction, ejection fraction, mass, and fractional area change were calculated. Percent difference was calculated as (interobservation difference/mean)×100. Interobserver reproducibility for both acquisitions was better for both volume and dimension measurements (P≤0.002) compared with area measurements, whereas intraobserver, interacquisition (for both observers), and interobserver-acquisition reproducibilities (for both observer-acquisition sets) were best for volume measurements (P≤0.01). Overall, interobserver-acquisition percent differences were significantly higher than interobserver and interacquisition percent differences (P<0.001).

CONCLUSIONS

In pediatric patients with dilated cardiomyopathy, compared with dimension and area methods, left ventricular measurements by volume method have the best reproducibility in settings where assessment is not performed by the same personnel.

CLINICAL TRIAL REGISTRATION

URL: https://www.clinicaltrials.gov. Unique identifier: NCT00123071.

Validity of automated measurement of left ventricular ejection fraction and volume using the Philips EPIQ system

PURPOSE

To assess the efficiency and reproducibility of automated measurements of left ventricular (LV) volumes and LV ejection fraction (LVEF) in comparison to manually traced biplane Simpson's method.

METHOD

This is a single-center prospective study. Apical four- and two-chamber views were acquired in patients in sinus rhythm. Two operators independently measured LV volumes and LVEF using biplane Simpson's method. In addition, the image analysis software a2DQ on the Philips EPIQ system was applied to automatically assess the LV volumes and LVEF. Time spent on each analysis, using both methods, was documented. Concordance of echocardiographic measures was evaluated using intraclass correlation (ICC) and Bland-Altman analysis.

RESULTS

Manual tracing and automated measurement of LV volumes and LVEF were performed in 184 patients with a mean age of 67.3 ± 17.3 years and BMI 28.0 ± 6.8 kg/m² . ICC and Bland-Altman analysis showed good agreements between manual and automated methods measuring LVEF, end-systolic, and end-diastolic volumes. The average analysis time was significantly less using the automated method than manual tracing (116 vs 217 seconds/patient, P < .0001).

CONCLUSION

Automated measurement using the novel image analysis software a2DQ on the Philips EPIQ system produced accurate, efficient, and reproducible assessment of LV volumes and LVEF compared with manual measurement.

Variability in echocardiographic measurements of left ventricular function in septic shock patients

Background

Echocardiography is increasingly used for haemodynamic evaluation and titration of therapy in intensive care, warranting reliable and reproducible measurements. The aim of this study was to evaluate the observer dependence of echocardiographic findings of left ventricular (LV) diastolic and systolic dysfunction in patients with septic shock.

Methods

Echocardiograms performed in 47 adult patients admitted with septic shock to a general intensive care unit (ICU) were independently evaluated by one cardiologist and one intensivist for the following signs: decreased diastolic tissue velocity of the base of the LV septum (é), increased early mitral inflow (E) to é ratio (E/é), decreased LV ejection fraction (EF) and decreased LV global longitudinal peak strain (GLPS). Diastolic dysfunction was defined as é <8.0 cm/s and/or E/é ≥15 and systolic dysfunction as EF <50% and/or GLPS > −15%. Ten randomly selected examinations were re-analysed two months later. Pearson’s r was used to test the correlation and Bland-Altman plots to assess the agreement between observers. Kappa statistics were used to test the consistency between readers and intraclass correlation coefficients (ICC) for inter- and intraobserver variability.

Results

In 44 patients (94%), image quality was sufficient for echocardiographic measurements. The agreement between observers was moderate (k = 0.60 for é, k = 0.50 for E/é and k = 0.60 for EF) to good (k = 0.71 for GLPS). Pearson’s r was 0.76 for é, 0.85 for E/é, 0.78 for EF and 0.84 for GLPS (p < 0.001 for all four). The ICC between observers for é was very good (0.85; 95% confidence interval (CI) 0.73-0.92), good for E/é (0.70; 95% CI 0.45 – 0.84), very good for EF (0.87; 95% CI 0.77 – 0.93), excellent for GLPS (0.91; 95% CI 0.74 – 0.95), and very good for all measures repeated by one of the observers. On Bland-Altman analysis, the mean differences and 95% limits of agreement for é, E/é, EF and GLPS were −0.01 (0.04 – 0.07), 2.0 (−14.2 – 18.1), 0.86 (−16 – 14.3) and 0.04 (−5.04 – 5.12), respectively.

Conclusions

Moderate observer-related differences in assessing LV dysfunction were seen. GLPS is the least user dependent and most reproducible echocardiographic measurement of LV function in septic shock.

Electronic supplementary material

The online version of this article (doi:10.1186/s12947-015-0015-6) contains supplementary material, which is available to authorized users.

Effect of transendocardial delivery of autologous bone marrow mononuclear cells on functional capacity, left ventricular function, and perfusion in chronic heart failure: the FOCUS-CCTRN trial

CONTEXT

Previous studies using autologous bone marrow mononuclear cells (BMCs) in patients with ischemic cardiomyopathy have demonstrated safety and suggested efficacy.

OBJECTIVE

To determine if administration of BMCs through transendocardial injections improves myocardial perfusion, reduces left ventricular end-systolic volume (LVESV), or enhances maximal oxygen consumption in patients with coronary artery disease or LV dysfunction, and limiting heart failure or angina.

DESIGN, SETTING, AND PATIENTS

A phase 2 randomized double-blind, placebo-controlled trial of symptomatic patients (New York Heart Association classification II-III or Canadian Cardiovascular Society classification II-IV) with a left ventricular ejection fraction of 45% or less, a perfusion defect by single-photon emission tomography (SPECT), and coronary artery disease not amenable to revascularization who were receiving maximal medical therapy at 5 National Heart, Lung, and Blood Institute-sponsored Cardiovascular Cell Therapy Research Network (CCTRN) sites between April 29, 2009, and April 18, 2011.

INTERVENTION

Bone marrow aspiration (isolation of BMCs using a standardized automated system performed locally) and transendocardial injection of 100 million BMCs or placebo (ratio of 2 for BMC group to 1 for placebo group).

MAIN OUTCOME MEASURES

Co-primary end points assessed at 6 months: changes in LVESV assessed by echocardiography, maximal oxygen consumption, and reversibility on SPECT. Phenotypic and functional analyses of the cell product were performed by the CCTRN biorepository core laboratory.

RESULTS

Of 153 patients who provided consent, a total of 92 (82 men; average age: 63 years) were randomized (n = 61 in BMC group and n = 31 in placebo group). Changes in LVESV index (-0.9 mL/m(2) [95% CI, -6.1 to 4.3]; P = .73), maximal oxygen consumption (1.0 [95% CI, -0.42 to 2.34]; P = .17), and reversible defect (-1.2 [95% CI, -12.50 to 10.12]; P = .84) were not statistically significant. There were no differences found in any of the secondary outcomes, including percent myocardial defect, total defect size, fixed defect size, regional wall motion, and clinical improvement.

CONCLUSION

Among patients with chronic ischemic heart failure, transendocardial injection of autologous BMCs compared with placebo did not improve LVESV, maximal oxygen consumption, or reversibility on SPECT.

TRIAL REGISTRATION

clinicaltrials.gov Identifier: NCT00824005.